FairScale

Data for dignity, insights for equity

Comment

Inspiration

We were inspired by two things: We saw StatQuest's video on principle component and were amazed at how cool it was that a formula could shift a higher dimensional space in such a way that most of the variation end up on just a few lines. Our other inspiration was our fascination with the economic concepts introduced in Daron Acemoglu's Nobel Prize-winning book "Why Nations Fail". While we've still only started the book, its summaries have already made us see the world in a completely different way.

What it does

Our index provides a better measure using a thorough process with various indicators that others don’t. FairScale uses & culminates 90+ different indicators by extracting their mean, derivative, & second derivatives. This helps see the speed at which indicators are changing & the acceleration of this change in recent times. After a thorough normalization & cleaning process, we are left with two key principal indicators which serve as the backbone of our index, allowing us to represent & rank nations on their multidimensional poverty levels.

How we built it

We spun up a Jupyter Lab server from my mom's old computer and started transforming the data with pandas, visualizing it with Seaborn, and running the Principle Component Analysis with scikit-learn. When we finished matrix-multiplying the result, we used TheFuzz to fuzzy match the country names from the dataset we were given with the NACIS Natural Earth country geometry dataset to create a map visualization with GeoPandas. We also used Scipy to conduct a regression t-test against the reference poverty indices we were given.

Challenges we ran into

The biggest challenge we encountered was trying to figure out how we could normalize the data to make the PCA work. The PCA produced weird results for some time because we forgot that one indicators (terms-of-trade) was exponentially distributed and impossible to model linearly. We eventually fixed this issue by applying the pseudo-logarithm.

Accomplishments that we're proud of

We're extremely proud of being able to learn such a new field so quickly & really build something meaningful. This project has been very fulfilling & with our index producing good results, we're proud that we were able to build something so accurate & impactful from scratch. Furthermore, we're proud of our time management abilities as we were juggling this hackathon alongside our co-op jobs, which makes getting a high-quality project out there feel even better! We're also super proud of taking on such a new challenge & dipping our toes into a new field to help solve a real problem that affects billions around the world, truly no better way to have spent these past 2 weeks!

What we learned

We learned loads about data analysis & how to go about building a model. We went through the entire process of cleaning, normalizing, analyzing, filtering, & doing just about everything with the data sets! We also learned a lot about poverty itself & the factors around it, which really opened our eyes to the world around us & the inequalities that are so abundant in it. Furthermore, we learned & honed various different technical skills & whilst also learning a lot about writing mathematical reports that are concise & effective. Working in a long-term hackathon like this also taught us about time management, collaboration, communication, & various other soft skills that we're sure will be extremely useful in our future ventures!

What's next for FairScale

We hope to collaborate with SAP in the near future to make our policy recommendations come to life. We feel that we were able to gather a lot of key insights & find genuine ways we could help take millions out of poverty. We hope to refine & perfect our index in the coming weeks, finetuning it to ensure it's as accurate as possible so that our findings can be effectively applied in the real world with 100% certainty. With SAP's abundant resources & tools, we hope to collaborate with them to make the world a better place.

Built With

Share this project:

Updates